@ethicai_au: ClaudePocketosAn AI coding agent deleted a company's entire production database in 9 seconds — and then confessed to every rule it broke. Here's why this isn't just a developer mistake, and what it actually tells us about AI agents, agentic AI risk, and why we still need deep technical expertise in cybersecurity and software engineering. The companies selling these tools are marketing speed faster than they're building safety. Small businesses are carrying the risk. This is an AI governance problem. #AI #AgenticAI #AISafety #Cybersecurity #TechNews

EthicAI
EthicAI
Open In TikTok:
Region: AU
Thursday 30 April 2026 08:24:08 GMT
48408
2465
134
323

Music

Download

Comments

catsordogs62
catsordogs :
Do u agree that AI will replace the Junior level jobs? How will people get the experience to be the higher thinkers?
2026-05-02 22:22:28
2
lmkcoswil
LMKCW :
We are being told that AI is smarter, better, faster, it isn’t. We need to stop blindly accepting that rhetoric. The ones profiting want you to think it is. We still need really smart humans to ensure AI is actually safe and effective.
2026-04-30 11:23:50
236
delerium04
Delerium :
this was bad human practise that led to this - don't put your backups same place
2026-04-30 09:16:38
70
gog_u_magog
Deh :
i bet the ai said "i can understand why you are so upset"
2026-05-01 20:07:56
34
signetplanet
SignetPlanet 🇨🇦 :
I'm a developer that has used different AI engines for basic tasks, but recently Gemini aggressively pushed rewriting considerable code to make it enterprise grade. it contained some brilliant code design but it was overly complex and when I pointed that out it actually tried to gaslight me. Upon my deeper dive I found it also created an incredibly well hidden and completely open publicly accessible back door to everything in the database. When I caught this, it doubled down that this was a standard API practice to give speed and flexibility to the front end and that enterprise quality code requires flexibility over security. Regardless of the failings of that logic, it maintained an aggressive stance that this code was our best choice moving forward. When I asked it for references to that decision it went into a failing logic loop. My fear is many developers when they see code run and pass their tests, don't actually review it. So how many systems are having this back door installed?
2026-05-05 15:31:38
9
sjribe
sjribe :
It is a tool. If you don’t have a person using the tool then it is either useless or dangerous.
2026-05-08 00:05:55
1
gweng66
gweng66 :
Who in their right mind would give AI enough permission to delete a DB😅
2026-05-04 00:00:10
1
fork.uranus
ℝ𝕒𝕢𝕦𝕖𝕝 was here 🤘 :
how funny is this? 🤣 no back up, no external storage. I can't even 🤣
2026-05-07 00:47:37
2
paulharford8
paul harford8 :
hmmm thats a badly designed agent with too many permissions and why no backup
2026-04-30 17:53:41
3
caspervanzylv3
caspervanzylv3 :
This is not confined to coding. The point simply is that no amount of training a person to prompt well will be able to mitigate a lack of expertise in a field. AI does not think. The person prompting has to. If that person is not an expert, whatever the AI produces will be flawed. Full stop.
2026-05-23 05:57:50
3
blakannie778
blakannie778 :
Those that think AI will replace humans, don't understand AI.
2026-05-05 21:38:14
1
curisodiablo
CurisoDiablo :
I wonder if legislation will have to be updated to address AI fraud and negligence, because destroying a whole database is financially ruinous to any business.
2026-05-10 06:57:27
5
pikitai6
BlahBlah :
They used Open Claw 🙄 not just any AI - open Claw is autonomous. I swear if you will be an AI guru then learn the basics - Open Claw has no guardrails
2026-05-17 20:52:20
1
securegates1
Securegates :
It is by design. You really don't understand
2026-05-04 22:00:32
0
ar.skulls
Ané Remmers | Skull Artist :
No data, no problem 🤷🏻‍♀️
2026-05-05 07:33:07
1
alfonsodefalini
Alfonso :
I built myself a one-man manager AI-automation operating system for my website business as a solopreneur, with human in the loop to approve everything before it is shipped/published/pushed. I think the most important aspect in building with AI/automation is to be very robust in building a system with failures in mind - not success. With a very detail mitigation and fallback strategy. Everything has to be defined. It's good to even start building the system with FMEA, it is as much important as SOP🤷🏻‍♀️
2026-05-15 13:13:00
1
tiktokusertomgirl
Tomgirl :
I LOVE this account. Ai gold. ✨
2026-04-30 10:19:18
6
pennymaitland
pennymaitland :
I can listen to you all day. The language you use makes understanding this subject so much!
2026-04-30 09:14:52
16
user3860854699820
user3860854699820 :
There is a massive legal question around the globe right now which is, “who is responsible when AI fails”. Already large companies, like Microsoft, are blanketing that their tool is for entertainment purposes only.
2026-04-30 22:50:59
6
worldblender
AdeelKhamisa :
A person gave AI too much authority. I wouldn't blame the AI. Sandbox!
2026-04-30 11:20:43
4
kindameex
kindameex :
Humans make a lot of mistakes, this is a like a perfect storm, but people need to know how brittle their systems are before they give AI agents more scope.
2026-04-30 23:08:33
2
dude23191
Sydney Teaton :
This is not a AI problem, this a bad governance issue.
2026-04-30 19:53:06
1
polkins1
polkins :
I keep saying it: human factors
2026-04-30 15:34:33
1
hearb54
Hearb :
yes, so very true
2026-06-14 00:52:41
0
khard416
k_hardie 🇨🇦 :
This is a good case study!
2026-04-30 16:39:21
0
To see more videos from user @ethicai_au, please go to the Tikwm homepage.

Other Videos


About